In this paper,we propose a learning algorithm termed linear multistep adaptive moment(LMAdam) to enhance the adaptive moment(Adam) algorithm for machine learning.Considering Adam as a single-step discretization of its...In this paper,we propose a learning algorithm termed linear multistep adaptive moment(LMAdam) to enhance the adaptive moment(Adam) algorithm for machine learning.Considering Adam as a single-step discretization of its continuous counterpart,we develop the LMAdam algorithm based on a linear multistep discretization scheme.We design a feedforward neural network for learning the coefficients of the multistep terms with ensured consistency and select the coefficients to ensure zero stability of the multistep terms.We experimentally demonstrate the superiority of the LMAdam via extensive experimentation on benchmark datasets for training various deep neural networks in three applications.展开更多
An effective algorithm is proposed to detect copy-move forgery.In this algorithm,first,the PatchMatch algorithm is improved by using a reliable order-statistics-based approximate nearest neighbor search algorithm(ROSA...An effective algorithm is proposed to detect copy-move forgery.In this algorithm,first,the PatchMatch algorithm is improved by using a reliable order-statistics-based approximate nearest neighbor search algorithm(ROSANNA)to modify the propagation process.Then,fractional quaternion Zernike moments(FrQZMs)are considered to be features extracted from color forged images.Finally,the extracted FrQZMs features are matched by the improved PatchMatch algorithm.The experimental results on two publicly available datasets(FAU and GRIP datasets)show that the proposed algorithm performs better than the state-of-the-art algorithms not only in objective criteria F-measure value but also in visual.Moreover,the proposed algorithm is robust to some attacks,such as additive white Gaussian noise,JPEG compression,rotation,and scaling.展开更多
【目的】针对现有桥梁施工线形预测方法的不足,提出一种基于自适应矩估计(adaptive moment estimation,Adam)优化反向传播(back propagation,BP)神经网络的连续刚构桥线形预测方法。【方法】以小乌江大桥为研究对象,通过正交试验确定了...【目的】针对现有桥梁施工线形预测方法的不足,提出一种基于自适应矩估计(adaptive moment estimation,Adam)优化反向传播(back propagation,BP)神经网络的连续刚构桥线形预测方法。【方法】以小乌江大桥为研究对象,通过正交试验确定了桥梁施工线形的敏感参数为混凝土容重、混凝土弹性模量、张拉控制应力和温度。以均方根误差、平均绝对误差、决定系数和运算耗时为评价指标,在初始学习率相同的条件下,对梯度下降、梯度下降最小化、均方根传播和Adam四种优化算法的性能进行对比。【结果】基于Adam优化算法的BP神经网络收敛时的运算耗时为0.518 s,相较于其他三种优化算法,Adam优化算法下BP神经网络具有更快的收敛速度和更高的拟合精度。【结论】所提方法可较准确地预测连续刚构桥施工过程的线形。展开更多
针对电力领域文本数据分词准确性较低的问题,提出一种基于改进ADAM(adaptive moment estimation)算法的中文分词技术。选用Skip-Gram模型作为字嵌入模型,将字词转为分布式向量,搭建卷积神经网络-门控循环单元-条件随机场(CNN-Bi-GRU-CRF...针对电力领域文本数据分词准确性较低的问题,提出一种基于改进ADAM(adaptive moment estimation)算法的中文分词技术。选用Skip-Gram模型作为字嵌入模型,将字词转为分布式向量,搭建卷积神经网络-门控循环单元-条件随机场(CNN-Bi-GRU-CRF)模型实现电力领域文本语句的分割,提出一种改进的ADAM算法,通过控制不同时间窗口的学习率优化神经网络模型,提高模型训练速度。将所提算法运用于变电站SCD(system configuration description)文本数据分词的算例分析,通过与其他主流分词算法进行比较,验证所提分词技术的先进性与准确性。展开更多
Magnetic tracking technologies have a promising application in detecting the real-time position andattitude of a capsule endoscope.However,most of them need to measure the magnetic moment of a permanentmagnet(PM)embed...Magnetic tracking technologies have a promising application in detecting the real-time position andattitude of a capsule endoscope.However,most of them need to measure the magnetic moment of a permanentmagnet(PM)embedded in the capsule accurately in advance,which can cause inconvenience to practical application.To solve this problem,this paper proposes a magnetic tracking system with the capability of measuring themagnetic moment of the PM automatically.The system is constructed based on a 4×4 magnetic sensor array,whose sensing data is analyzed to determine the magnetic moment by referring to a magnetic dipole model.Withthe determined magnetic moment,a method of fusing the linear calculation and Levenberg-Marquardt algorithmsis proposed to determine the 3D position and 2D attitude of the PM.The experiments verified that the proposedsystem can achieve localization errors of 0.48 mm,0.42 mm,and 0.83 mm and orientation errors of 0.66◦,0.64◦,and 0.87◦for a PM(∅10 mm×10 mm)at vertical heights of 5 cm,10 cm,and 15 cm from the magnetic sensorarray,respectively.展开更多
In this paper,we propose enhancements to Beetle Antennae search(BAS)algorithm,called BAS-ADAIVL to smoothen the convergence behavior and avoid trapping in localminima for a highly noin-convex objective function.We ach...In this paper,we propose enhancements to Beetle Antennae search(BAS)algorithm,called BAS-ADAIVL to smoothen the convergence behavior and avoid trapping in localminima for a highly noin-convex objective function.We achieve this by adaptively adjusting the step-size in each iteration using the adaptive moment estimation(ADAM)update rule.The proposed algorithm also increases the convergence rate in a narrow valley.A key feature of the ADAM update rule is the ability to adjust the step-size for each dimension separately instead of using the same step-size.Since ADAM is traditionally used with gradient-based optimization algorithms,therefore we first propose a gradient estimation model without the need to differentiate the objective function.Resultantly,it demonstrates excellent performance and fast convergence rate in searching for the optimum of noin-convex functions.The efficiency of the proposed algorithm was tested on three different benchmark problems,including the training of a high-dimensional neural network.The performance is compared with particle swarm optimizer(PSO)and the original BAS algorithm.展开更多
It is well-known that appropriately determining the solution space and initializations is crucial for obtaining high-quality optimal solutions for optimization problems,considering both gradient-based and gradient-fre...It is well-known that appropriately determining the solution space and initializations is crucial for obtaining high-quality optimal solutions for optimization problems,considering both gradient-based and gradient-free techniques.However,the general framework for adaptively dealing with a particular optimization problem is commonly overlooked and thus still hidden in the literature.To overcome this limitation,a new approach assisted by the neural network(NN)is proposed for solving high-dimensional optimization issues.By restructuring the search space to optimize the objective function via a nonlinear mapping constructed by an autoencoder(AE),the surrogate solution space is constructed by a network training process and dynamically oriented to the optimal solution of the optimization issue.To enhance the optimization efficiency and address non-smooth problems,the classical metaheuristic grey wolf optimizer(GWO)and the adaptive moment estimation(Adam)are sequentially employed to complement the disadvantages of the constituted models.The effectiveness of the proposed approach is validated by solving a set of mathematical functions with 1000-dimensional and three large-scale truss design optimization problems.Several numerical experiments show that the solution space is reduced in terms of both size and complexity based on the restructuring procedure,in which the global optimal solution is still conserved,leading to better optimization efficiency when solving optimization problems with complex search domains with large dimensions.In addition,the hybrid optimizer has also been proven to be more effective when combined with the restructuring technique owing to the use of the Adam algorithm in the second phase.展开更多
A novel algorithm is presented to make the results of image matching more reliable and accurate based on SIFT (Scale Invariant Feature Transform). SIFT algorithm has been identified as the most resistant matching algo...A novel algorithm is presented to make the results of image matching more reliable and accurate based on SIFT (Scale Invariant Feature Transform). SIFT algorithm has been identified as the most resistant matching algorithm to common image deformations; however, if there are similar regions in images, SIFT algorithm still generates some analogical descriptors and provides many mismatches. This paper examines the local image descriptor used by SIFT and presents a new algorithm by integrating SIFT with two-dimensional moment invariants and disparity gradient to improve the matching results. In the new algorithm, decision tree is used, and the whole matching process is divided into three levels with different primitives. Matching points are considered as correct ones only when they satisfy all the three similarity measurements. Experiment results demonstrate that the new approach is more reliable and accurate.展开更多
基金supported in part by the National Natural Science Foundation of China(62506148 and 62476115)the Fundamental Research Funds for the Central Universities(lzujbky-2025-pd05 and lzujbky-2025-ytB01)+2 种基金the Research Grants Council of the Hong Kong Special Administrative Region of China(AoE/E-407/24-N and C1013-24G)the Postdoctoral Fellowship Program(Grade C) of China Postdoctoral Science Foundation(GZC20251039)the Supercomputing Center of Lanzhou University。
摘要In this paper,we propose a learning algorithm termed linear multistep adaptive moment(LMAdam) to enhance the adaptive moment(Adam) algorithm for machine learning.Considering Adam as a single-step discretization of its continuous counterpart,we develop the LMAdam algorithm based on a linear multistep discretization scheme.We design a feedforward neural network for learning the coefficients of the multistep terms with ensured consistency and select the coefficients to ensure zero stability of the multistep terms.We experimentally demonstrate the superiority of the LMAdam via extensive experimentation on benchmark datasets for training various deep neural networks in three applications.
基金The National Natural Science of China(No.61572258,61771231,61772281,61672294)the Priority Academic Program Development of Jiangsu Higher Education Institutionsthe Qing Lan Project of Jiangsu Higher Education Institutions
摘要An effective algorithm is proposed to detect copy-move forgery.In this algorithm,first,the PatchMatch algorithm is improved by using a reliable order-statistics-based approximate nearest neighbor search algorithm(ROSANNA)to modify the propagation process.Then,fractional quaternion Zernike moments(FrQZMs)are considered to be features extracted from color forged images.Finally,the extracted FrQZMs features are matched by the improved PatchMatch algorithm.The experimental results on two publicly available datasets(FAU and GRIP datasets)show that the proposed algorithm performs better than the state-of-the-art algorithms not only in objective criteria F-measure value but also in visual.Moreover,the proposed algorithm is robust to some attacks,such as additive white Gaussian noise,JPEG compression,rotation,and scaling.
摘要【目的】针对现有桥梁施工线形预测方法的不足,提出一种基于自适应矩估计(adaptive moment estimation,Adam)优化反向传播(back propagation,BP)神经网络的连续刚构桥线形预测方法。【方法】以小乌江大桥为研究对象,通过正交试验确定了桥梁施工线形的敏感参数为混凝土容重、混凝土弹性模量、张拉控制应力和温度。以均方根误差、平均绝对误差、决定系数和运算耗时为评价指标,在初始学习率相同的条件下,对梯度下降、梯度下降最小化、均方根传播和Adam四种优化算法的性能进行对比。【结果】基于Adam优化算法的BP神经网络收敛时的运算耗时为0.518 s,相较于其他三种优化算法,Adam优化算法下BP神经网络具有更快的收敛速度和更高的拟合精度。【结论】所提方法可较准确地预测连续刚构桥施工过程的线形。
摘要针对电力领域文本数据分词准确性较低的问题,提出一种基于改进ADAM(adaptive moment estimation)算法的中文分词技术。选用Skip-Gram模型作为字嵌入模型,将字词转为分布式向量,搭建卷积神经网络-门控循环单元-条件随机场(CNN-Bi-GRU-CRF)模型实现电力领域文本语句的分割,提出一种改进的ADAM算法,通过控制不同时间窗口的学习率优化神经网络模型,提高模型训练速度。将所提算法运用于变电站SCD(system configuration description)文本数据分词的算例分析,通过与其他主流分词算法进行比较,验证所提分词技术的先进性与准确性。
基金the National Natural Science Foundation of China(Nos.52275038 and 61803347)the Shanxi Province Science Foundation for Excellent Youth(No.202203021224007)+1 种基金the Key Research and Development Plan of Shanxi Province(No.201903D321164)the Opening Foundation of Shanxi Key Laboratory of Advanced Manufacturing Technology(No.XJZZ202101)。
摘要Magnetic tracking technologies have a promising application in detecting the real-time position andattitude of a capsule endoscope.However,most of them need to measure the magnetic moment of a permanentmagnet(PM)embedded in the capsule accurately in advance,which can cause inconvenience to practical application.To solve this problem,this paper proposes a magnetic tracking system with the capability of measuring themagnetic moment of the PM automatically.The system is constructed based on a 4×4 magnetic sensor array,whose sensing data is analyzed to determine the magnetic moment by referring to a magnetic dipole model.Withthe determined magnetic moment,a method of fusing the linear calculation and Levenberg-Marquardt algorithmsis proposed to determine the 3D position and 2D attitude of the PM.The experiments verified that the proposedsystem can achieve localization errors of 0.48 mm,0.42 mm,and 0.83 mm and orientation errors of 0.66◦,0.64◦,and 0.87◦for a PM(∅10 mm×10 mm)at vertical heights of 5 cm,10 cm,and 15 cm from the magnetic sensorarray,respectively.
摘要In this paper,we propose enhancements to Beetle Antennae search(BAS)algorithm,called BAS-ADAIVL to smoothen the convergence behavior and avoid trapping in localminima for a highly noin-convex objective function.We achieve this by adaptively adjusting the step-size in each iteration using the adaptive moment estimation(ADAM)update rule.The proposed algorithm also increases the convergence rate in a narrow valley.A key feature of the ADAM update rule is the ability to adjust the step-size for each dimension separately instead of using the same step-size.Since ADAM is traditionally used with gradient-based optimization algorithms,therefore we first propose a gradient estimation model without the need to differentiate the objective function.Resultantly,it demonstrates excellent performance and fast convergence rate in searching for the optimum of noin-convex functions.The efficiency of the proposed algorithm was tested on three different benchmark problems,including the training of a high-dimensional neural network.The performance is compared with particle swarm optimizer(PSO)and the original BAS algorithm.
基金Project supported by the Basic Science Research Program through the National Research Foundation of Korea(NRF),funded by the Ministry of Science and ICT(No.RS-2024-00337001)。
摘要It is well-known that appropriately determining the solution space and initializations is crucial for obtaining high-quality optimal solutions for optimization problems,considering both gradient-based and gradient-free techniques.However,the general framework for adaptively dealing with a particular optimization problem is commonly overlooked and thus still hidden in the literature.To overcome this limitation,a new approach assisted by the neural network(NN)is proposed for solving high-dimensional optimization issues.By restructuring the search space to optimize the objective function via a nonlinear mapping constructed by an autoencoder(AE),the surrogate solution space is constructed by a network training process and dynamically oriented to the optimal solution of the optimization issue.To enhance the optimization efficiency and address non-smooth problems,the classical metaheuristic grey wolf optimizer(GWO)and the adaptive moment estimation(Adam)are sequentially employed to complement the disadvantages of the constituted models.The effectiveness of the proposed approach is validated by solving a set of mathematical functions with 1000-dimensional and three large-scale truss design optimization problems.Several numerical experiments show that the solution space is reduced in terms of both size and complexity based on the restructuring procedure,in which the global optimal solution is still conserved,leading to better optimization efficiency when solving optimization problems with complex search domains with large dimensions.In addition,the hybrid optimizer has also been proven to be more effective when combined with the restructuring technique owing to the use of the Adam algorithm in the second phase.
摘要A novel algorithm is presented to make the results of image matching more reliable and accurate based on SIFT (Scale Invariant Feature Transform). SIFT algorithm has been identified as the most resistant matching algorithm to common image deformations; however, if there are similar regions in images, SIFT algorithm still generates some analogical descriptors and provides many mismatches. This paper examines the local image descriptor used by SIFT and presents a new algorithm by integrating SIFT with two-dimensional moment invariants and disparity gradient to improve the matching results. In the new algorithm, decision tree is used, and the whole matching process is divided into three levels with different primitives. Matching points are considered as correct ones only when they satisfy all the three similarity measurements. Experiment results demonstrate that the new approach is more reliable and accurate.